Re: [PATCH 3/7] NetLabel: CIPSOv4 engine

[Paul Moore <paul.moore@hp.com>]

On Saturday 15 July 2006 9:15 am, James Morris wrote:
> On Fri, 14 Jul 2006, paul.moore@hp.com wrote:
> > +int cipso_v4_doi_add(struct cipso_v4_doi *doi_def)
> > +{
> > +	if (doi_def == NULL || doi_def->doi == CIPSO_V4_DOI_UNKNOWN)
> > +		return -EINVAL;
> > +
> > +	doi_def->valid = 1;
> > +	INIT_RCU_HEAD(&doi_def->rcu);
> > +	INIT_LIST_HEAD(&doi_def->dom_list);
> > +
> > +	rcu_read_lock();
> > +	if (cipso_v4_doi_search(doi_def->doi) != NULL) {
> > +		rcu_read_unlock();
> > +		return -EEXIST;
> > +	}
> > +	spin_lock(&cipso_v4_doi_list_lock);
> > +	list_add_tail_rcu(&doi_def->list, &cipso_v4_doi_list);
> > +	spin_unlock(&cipso_v4_doi_list_lock);
> > +	rcu_read_unlock();
> > +
> > +	return 0;
> > +}
>
> This is racy in that you can end up with duplicate entries in the list
> between rcu_read_lock() and spin_lock().
>
> You need to perform the search a second time under the spin_lock() as a
> slow path to verify that an entry wasn't added, before adding a new one.
>

Thanks, there are a few more place which might have similar problems - I'll 
check those as well.

-- 
paul moore
linux security @ hp